projects / ceph.git / blame
| Commit | Line | Data |
|---|---|---|
| 9f95a23c TL |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
| 2 | // vim: ts=8 sw=2 smarttab | |
| 3 | ||
| 4 | #include "osd_perf_counters.h" | |
| 5 | #include "include/common_fwd.h" | |
| 6 | ||
| 7 | ||
| 8 | PerfCounters *build_osd_logger(CephContext *cct) { | |
| 9 | PerfCountersBuilder osd_plb(cct, "osd", l_osd_first, l_osd_last); | |
| 10 | ||
| 11 | // Latency axis configuration for op histograms, values are in nanoseconds | |
| 12 | PerfHistogramCommon::axis_config_d op_hist_x_axis_config{ | |
| 13 | "Latency (usec)", | |
| 14 | PerfHistogramCommon::SCALE_LOG2, ///< Latency in logarithmic scale | |
| 15 | 0, ///< Start at 0 | |
| 16 | 100000, ///< Quantization unit is 100usec | |
| 17 | 32, ///< Enough to cover much longer than slow requests | |
| 18 | }; | |
| 19 | ||
| 20 | // Op size axis configuration for op histograms, values are in bytes | |
| 21 | PerfHistogramCommon::axis_config_d op_hist_y_axis_config{ | |
| 22 | "Request size (bytes)", | |
| 23 | PerfHistogramCommon::SCALE_LOG2, ///< Request size in logarithmic scale | |
| 24 | 0, ///< Start at 0 | |
| 25 | 512, ///< Quantization unit is 512 bytes | |
| 26 | 32, ///< Enough to cover requests larger than GB | |
| 27 | }; | |
| 28 | ||
| 29 | ||
| 30 | // All the basic OSD operation stats are to be considered useful | |
| 31 | osd_plb.set_prio_default(PerfCountersBuilder::PRIO_USEFUL); | |
| 32 | ||
| 33 | osd_plb.add_u64( | |
| 34 | l_osd_op_wip, "op_wip", | |
| 35 | "Replication operations currently being processed (primary)"); | |
| 36 | osd_plb.add_u64_counter( | |
| 37 | l_osd_op, "op", | |
| 38 | "Client operations", | |
| 39 | "ops", PerfCountersBuilder::PRIO_CRITICAL); | |
| 40 | osd_plb.add_u64_counter( | |
| 41 | l_osd_op_inb, "op_in_bytes", | |
| 42 | "Client operations total write size", | |
| 43 | "wr", PerfCountersBuilder::PRIO_INTERESTING, unit_t(UNIT_BYTES)); | |
| 44 | osd_plb.add_u64_counter( | |
| 45 | l_osd_op_outb, "op_out_bytes", | |
| 46 | "Client operations total read size", | |
| 47 | "rd", PerfCountersBuilder::PRIO_INTERESTING, unit_t(UNIT_BYTES)); | |
| 48 | osd_plb.add_time_avg( | |
| 49 | l_osd_op_lat, "op_latency", | |
| 50 | "Latency of client operations (including queue time)", | |
| 51 | "l", 9); | |
| 52 | osd_plb.add_time_avg( | |
| 53 | l_osd_op_process_lat, "op_process_latency", | |
| 54 | "Latency of client operations (excluding queue time)"); | |
| 55 | osd_plb.add_time_avg( | |
| 56 | l_osd_op_prepare_lat, "op_prepare_latency", | |
| 57 | "Latency of client operations (excluding queue time and wait for finished)"); | |
| 58 | ||
| 59 | osd_plb.add_u64_counter( | |
| 60 | l_osd_op_r, "op_r", "Client read operations"); | |
| 61 | osd_plb.add_u64_counter( | |
| 62 | l_osd_op_r_outb, "op_r_out_bytes", "Client data read", NULL, PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
| 63 | osd_plb.add_time_avg( | |
| 64 | l_osd_op_r_lat, "op_r_latency", | |
| 65 | "Latency of read operation (including queue time)"); | |
| 66 | osd_plb.add_u64_counter_histogram( | |
| 67 | l_osd_op_r_lat_outb_hist, "op_r_latency_out_bytes_histogram", | |
| 68 | op_hist_x_axis_config, op_hist_y_axis_config, | |
| 69 | "Histogram of operation latency (including queue time) + data read"); | |
| 70 | osd_plb.add_time_avg( | |
| 71 | l_osd_op_r_process_lat, "op_r_process_latency", | |
| 72 | "Latency of read operation (excluding queue time)"); | |
| 73 | osd_plb.add_time_avg( | |
| 74 | l_osd_op_r_prepare_lat, "op_r_prepare_latency", | |
| 75 | "Latency of read operations (excluding queue time and wait for finished)"); | |
| 76 | osd_plb.add_u64_counter( | |
| 77 | l_osd_op_w, "op_w", "Client write operations"); | |
| 78 | osd_plb.add_u64_counter( | |
| 79 | l_osd_op_w_inb, "op_w_in_bytes", "Client data written"); | |
| 80 | osd_plb.add_time_avg( | |
| 81 | l_osd_op_w_lat, "op_w_latency", | |
| 82 | "Latency of write operation (including queue time)"); | |
| 83 | osd_plb.add_u64_counter_histogram( | |
| 84 | l_osd_op_w_lat_inb_hist, "op_w_latency_in_bytes_histogram", | |
| 85 | op_hist_x_axis_config, op_hist_y_axis_config, | |
| 86 | "Histogram of operation latency (including queue time) + data written"); | |
| 87 | osd_plb.add_time_avg( | |
| 88 | l_osd_op_w_process_lat, "op_w_process_latency", | |
| 89 | "Latency of write operation (excluding queue time)"); | |
| 90 | osd_plb.add_time_avg( | |
| 91 | l_osd_op_w_prepare_lat, "op_w_prepare_latency", | |
| 92 | "Latency of write operations (excluding queue time and wait for finished)"); | |
| 93 | osd_plb.add_u64_counter( | |
| 94 | l_osd_op_rw, "op_rw", | |
| 95 | "Client read-modify-write operations"); | |
| 96 | osd_plb.add_u64_counter( | |
| 97 | l_osd_op_rw_inb, "op_rw_in_bytes", | |
| 98 | "Client read-modify-write operations write in", NULL, PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
| 99 | osd_plb.add_u64_counter( | |
| 100 | l_osd_op_rw_outb,"op_rw_out_bytes", | |
| 101 | "Client read-modify-write operations read out ", NULL, PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
| 102 | osd_plb.add_time_avg( | |
| 103 | l_osd_op_rw_lat, "op_rw_latency", | |
| 104 | "Latency of read-modify-write operation (including queue time)"); | |
| 105 | osd_plb.add_u64_counter_histogram( | |
| 106 | l_osd_op_rw_lat_inb_hist, "op_rw_latency_in_bytes_histogram", | |
| 107 | op_hist_x_axis_config, op_hist_y_axis_config, | |
| 108 | "Histogram of rw operation latency (including queue time) + data written"); | |
| 109 | osd_plb.add_u64_counter_histogram( | |
| 110 | l_osd_op_rw_lat_outb_hist, "op_rw_latency_out_bytes_histogram", | |
| 111 | op_hist_x_axis_config, op_hist_y_axis_config, | |
| 112 | "Histogram of rw operation latency (including queue time) + data read"); | |
| 113 | osd_plb.add_time_avg( | |
| 114 | l_osd_op_rw_process_lat, "op_rw_process_latency", | |
| 115 | "Latency of read-modify-write operation (excluding queue time)"); | |
| 116 | osd_plb.add_time_avg( | |
| 117 | l_osd_op_rw_prepare_lat, "op_rw_prepare_latency", | |
| 118 | "Latency of read-modify-write operations (excluding queue time and wait for finished)"); | |
| 119 | ||
| 120 | // Now we move on to some more obscure stats, revert to assuming things | |
| 121 | // are low priority unless otherwise specified. | |
| 122 | osd_plb.set_prio_default(PerfCountersBuilder::PRIO_DEBUGONLY); | |
| 123 | ||
| 124 | osd_plb.add_time_avg(l_osd_op_before_queue_op_lat, "op_before_queue_op_lat", | |
| 125 | "Latency of IO before calling queue(before really queue into ShardedOpWq)"); // client io before queue op_wq latency | |
| 126 | osd_plb.add_time_avg(l_osd_op_before_dequeue_op_lat, "op_before_dequeue_op_lat", | |
| 127 | "Latency of IO before calling dequeue_op(already dequeued and get PG lock)"); // client io before dequeue_op latency | |
| 128 | ||
| 129 | osd_plb.add_u64_counter( | |
| 130 | l_osd_sop, "subop", "Suboperations"); | |
| 131 | osd_plb.add_u64_counter( | |
| 132 | l_osd_sop_inb, "subop_in_bytes", "Suboperations total size", NULL, 0, unit_t(UNIT_BYTES)); | |
| 133 | osd_plb.add_time_avg(l_osd_sop_lat, "subop_latency", "Suboperations latency"); | |
| 134 | ||
| 135 | osd_plb.add_u64_counter(l_osd_sop_w, "subop_w", "Replicated writes"); | |
| 136 | osd_plb.add_u64_counter( | |
| 137 | l_osd_sop_w_inb, "subop_w_in_bytes", "Replicated written data size", NULL, 0, unit_t(UNIT_BYTES)); | |
| 138 | osd_plb.add_time_avg( | |
| 139 | l_osd_sop_w_lat, "subop_w_latency", "Replicated writes latency"); | |
| 140 | osd_plb.add_u64_counter( | |
| 141 | l_osd_sop_pull, "subop_pull", "Suboperations pull requests"); | |
| 142 | osd_plb.add_time_avg( | |
| 143 | l_osd_sop_pull_lat, "subop_pull_latency", "Suboperations pull latency"); | |
| 144 | osd_plb.add_u64_counter( | |
| 145 | l_osd_sop_push, "subop_push", "Suboperations push messages"); | |
| 146 | osd_plb.add_u64_counter( | |
| 147 | l_osd_sop_push_inb, "subop_push_in_bytes", "Suboperations pushed size", NULL, 0, unit_t(UNIT_BYTES)); | |
| 148 | osd_plb.add_time_avg( | |
| 149 | l_osd_sop_push_lat, "subop_push_latency", "Suboperations push latency"); | |
| 150 | ||
| 151 | osd_plb.add_u64_counter(l_osd_pull, "pull", "Pull requests sent"); | |
| 152 | osd_plb.add_u64_counter(l_osd_push, "push", "Push messages sent"); | |
| 153 | osd_plb.add_u64_counter(l_osd_push_outb, "push_out_bytes", "Pushed size", NULL, 0, unit_t(UNIT_BYTES)); | |
| 154 | ||
| 155 | osd_plb.add_u64_counter( | |
| 156 | l_osd_rop, "recovery_ops", | |
| 157 | "Started recovery operations", | |
| 158 | "rop", PerfCountersBuilder::PRIO_INTERESTING); | |
| 159 | ||
| 160 | osd_plb.add_u64_counter( | |
| 161 | l_osd_rbytes, "recovery_bytes", | |
| 162 | "recovery bytes", | |
| 163 | "rbt", PerfCountersBuilder::PRIO_INTERESTING); | |
| 164 | ||
| 165 | osd_plb.add_u64(l_osd_loadavg, "loadavg", "CPU load"); | |
| 166 | osd_plb.add_u64( | |
| 167 | l_osd_cached_crc, "cached_crc", "Total number getting crc from crc_cache"); | |
| 168 | osd_plb.add_u64( | |
| 169 | l_osd_cached_crc_adjusted, "cached_crc_adjusted", | |
| 170 | "Total number getting crc from crc_cache with adjusting"); | |
| 171 | osd_plb.add_u64(l_osd_missed_crc, "missed_crc", | |
| 172 | "Total number of crc cache misses"); | |
| 173 | ||
| 174 | osd_plb.add_u64(l_osd_pg, "numpg", "Placement groups", | |
| 175 | "pgs", PerfCountersBuilder::PRIO_USEFUL); | |
| 176 | osd_plb.add_u64( | |
| 177 | l_osd_pg_primary, "numpg_primary", | |
| 178 | "Placement groups for which this osd is primary"); | |
| 179 | osd_plb.add_u64( | |
| 180 | l_osd_pg_replica, "numpg_replica", | |
| 181 | "Placement groups for which this osd is replica"); | |
| 182 | osd_plb.add_u64( | |
| 183 | l_osd_pg_stray, "numpg_stray", | |
| 184 | "Placement groups ready to be deleted from this osd"); | |
| 185 | osd_plb.add_u64( | |
| 186 | l_osd_pg_removing, "numpg_removing", | |
| 187 | "Placement groups queued for local deletion", "pgsr", | |
| 188 | PerfCountersBuilder::PRIO_USEFUL); | |
| 189 | osd_plb.add_u64( | |
| 190 | l_osd_hb_to, "heartbeat_to_peers", "Heartbeat (ping) peers we send to"); | |
| 191 | osd_plb.add_u64_counter(l_osd_map, "map_messages", "OSD map messages"); | |
| 192 | osd_plb.add_u64_counter(l_osd_mape, "map_message_epochs", "OSD map epochs"); | |
| 193 | osd_plb.add_u64_counter( | |
| 194 | l_osd_mape_dup, "map_message_epoch_dups", "OSD map duplicates"); | |
| 195 | osd_plb.add_u64_counter( | |
| 196 | l_osd_waiting_for_map, "messages_delayed_for_map", | |
| 197 | "Operations waiting for OSD map"); | |
| 198 | ||
| 199 | osd_plb.add_u64_counter( | |
| 200 | l_osd_map_cache_hit, "osd_map_cache_hit", "osdmap cache hit"); | |
| 201 | osd_plb.add_u64_counter( | |
| 202 | l_osd_map_cache_miss, "osd_map_cache_miss", "osdmap cache miss"); | |
| 203 | osd_plb.add_u64_counter( | |
| 204 | l_osd_map_cache_miss_low, "osd_map_cache_miss_low", | |
| 205 | "osdmap cache miss below cache lower bound"); | |
| 206 | osd_plb.add_u64_avg( | |
| 207 | l_osd_map_cache_miss_low_avg, "osd_map_cache_miss_low_avg", | |
| 208 | "osdmap cache miss, avg distance below cache lower bound"); | |
| 209 | osd_plb.add_u64_counter( | |
| 210 | l_osd_map_bl_cache_hit, "osd_map_bl_cache_hit", | |
| 211 | "OSDMap buffer cache hits"); | |
| 212 | osd_plb.add_u64_counter( | |
| 213 | l_osd_map_bl_cache_miss, "osd_map_bl_cache_miss", | |
| 214 | "OSDMap buffer cache misses"); | |
| 215 | ||
| 216 | osd_plb.add_u64( | |
| 217 | l_osd_stat_bytes, "stat_bytes", "OSD size", "size", | |
| 218 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
| 219 | osd_plb.add_u64( | |
| 220 | l_osd_stat_bytes_used, "stat_bytes_used", "Used space", "used", | |
| 221 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
| 222 | osd_plb.add_u64(l_osd_stat_bytes_avail, "stat_bytes_avail", "Available space", NULL, 0, unit_t(UNIT_BYTES)); | |
| 223 | ||
| 224 | osd_plb.add_u64_counter( | |
| 225 | l_osd_copyfrom, "copyfrom", "Rados \"copy-from\" operations"); | |
| 226 | ||
| 227 | osd_plb.add_u64_counter(l_osd_tier_promote, "tier_promote", "Tier promotions"); | |
| 228 | osd_plb.add_u64_counter(l_osd_tier_flush, "tier_flush", "Tier flushes"); | |
| 229 | osd_plb.add_u64_counter( | |
| 230 | l_osd_tier_flush_fail, "tier_flush_fail", "Failed tier flushes"); | |
| 231 | osd_plb.add_u64_counter( | |
| 232 | l_osd_tier_try_flush, "tier_try_flush", "Tier flush attempts"); | |
| 233 | osd_plb.add_u64_counter( | |
| 234 | l_osd_tier_try_flush_fail, "tier_try_flush_fail", | |
| 235 | "Failed tier flush attempts"); | |
| 236 | osd_plb.add_u64_counter( | |
| 237 | l_osd_tier_evict, "tier_evict", "Tier evictions"); | |
| 238 | osd_plb.add_u64_counter( | |
| 239 | l_osd_tier_whiteout, "tier_whiteout", "Tier whiteouts"); | |
| 240 | osd_plb.add_u64_counter( | |
| 241 | l_osd_tier_dirty, "tier_dirty", "Dirty tier flag set"); | |
| 242 | osd_plb.add_u64_counter( | |
| 243 | l_osd_tier_clean, "tier_clean", "Dirty tier flag cleaned"); | |
| 244 | osd_plb.add_u64_counter( | |
| 245 | l_osd_tier_delay, "tier_delay", "Tier delays (agent waiting)"); | |
| 246 | osd_plb.add_u64_counter( | |
| 247 | l_osd_tier_proxy_read, "tier_proxy_read", "Tier proxy reads"); | |
| 248 | osd_plb.add_u64_counter( | |
| 249 | l_osd_tier_proxy_write, "tier_proxy_write", "Tier proxy writes"); | |
| 250 | ||
| 251 | osd_plb.add_u64_counter( | |
| 252 | l_osd_agent_wake, "agent_wake", "Tiering agent wake up"); | |
| 253 | osd_plb.add_u64_counter( | |
| 254 | l_osd_agent_skip, "agent_skip", "Objects skipped by agent"); | |
| 255 | osd_plb.add_u64_counter( | |
| 256 | l_osd_agent_flush, "agent_flush", "Tiering agent flushes"); | |
| 257 | osd_plb.add_u64_counter( | |
| 258 | l_osd_agent_evict, "agent_evict", "Tiering agent evictions"); | |
| 259 | ||
| 260 | osd_plb.add_u64_counter( | |
| 261 | l_osd_object_ctx_cache_hit, "object_ctx_cache_hit", "Object context cache hits"); | |
| 262 | osd_plb.add_u64_counter( | |
| 263 | l_osd_object_ctx_cache_total, "object_ctx_cache_total", "Object context cache lookups"); | |
| 264 | ||
| 265 | osd_plb.add_u64_counter(l_osd_op_cache_hit, "op_cache_hit"); | |
| 266 | osd_plb.add_time_avg( | |
| 267 | l_osd_tier_flush_lat, "osd_tier_flush_lat", "Object flush latency"); | |
| 268 | osd_plb.add_time_avg( | |
| 269 | l_osd_tier_promote_lat, "osd_tier_promote_lat", "Object promote latency"); | |
| 270 | osd_plb.add_time_avg( | |
| 271 | l_osd_tier_r_lat, "osd_tier_r_lat", "Object proxy read latency"); | |
| 272 | ||
| 273 | osd_plb.add_u64_counter( | |
| 274 | l_osd_pg_info, "osd_pg_info", "PG updated its info (using any method)"); | |
| 275 | osd_plb.add_u64_counter( | |
| 276 | l_osd_pg_fastinfo, "osd_pg_fastinfo", | |
| 277 | "PG updated its info using fastinfo attr"); | |
| 278 | osd_plb.add_u64_counter( | |
| 279 | l_osd_pg_biginfo, "osd_pg_biginfo", "PG updated its biginfo attr"); | |
| 280 | ||
| 281 | return osd_plb.create_perf_counters(); | |
| 282 | } | |
| 283 | ||
| 284 | ||
| 285 | PerfCounters *build_recoverystate_perf(CephContext *cct) { | |
| 286 | PerfCountersBuilder rs_perf(cct, "recoverystate_perf", rs_first, rs_last); | |
| 287 | ||
| 288 | rs_perf.add_time_avg(rs_initial_latency, "initial_latency", "Initial recovery state latency"); | |
| 289 | rs_perf.add_time_avg(rs_started_latency, "started_latency", "Started recovery state latency"); | |
| 290 | rs_perf.add_time_avg(rs_reset_latency, "reset_latency", "Reset recovery state latency"); | |
| 291 | rs_perf.add_time_avg(rs_start_latency, "start_latency", "Start recovery state latency"); | |
| 292 | rs_perf.add_time_avg(rs_primary_latency, "primary_latency", "Primary recovery state latency"); | |
| 293 | rs_perf.add_time_avg(rs_peering_latency, "peering_latency", "Peering recovery state latency"); | |
| 294 | rs_perf.add_time_avg(rs_backfilling_latency, "backfilling_latency", "Backfilling recovery state latency"); | |
| 295 | rs_perf.add_time_avg(rs_waitremotebackfillreserved_latency, "waitremotebackfillreserved_latency", "Wait remote backfill reserved recovery state latency"); | |
| 296 | rs_perf.add_time_avg(rs_waitlocalbackfillreserved_latency, "waitlocalbackfillreserved_latency", "Wait local backfill reserved recovery state latency"); | |
| 297 | rs_perf.add_time_avg(rs_notbackfilling_latency, "notbackfilling_latency", "Notbackfilling recovery state latency"); | |
| 298 | rs_perf.add_time_avg(rs_repnotrecovering_latency, "repnotrecovering_latency", "Repnotrecovering recovery state latency"); | |
| 299 | rs_perf.add_time_avg(rs_repwaitrecoveryreserved_latency, "repwaitrecoveryreserved_latency", "Rep wait recovery reserved recovery state latency"); | |
| 300 | rs_perf.add_time_avg(rs_repwaitbackfillreserved_latency, "repwaitbackfillreserved_latency", "Rep wait backfill reserved recovery state latency"); | |
| 301 | rs_perf.add_time_avg(rs_reprecovering_latency, "reprecovering_latency", "RepRecovering recovery state latency"); | |
| 302 | rs_perf.add_time_avg(rs_activating_latency, "activating_latency", "Activating recovery state latency"); | |
| 303 | rs_perf.add_time_avg(rs_waitlocalrecoveryreserved_latency, "waitlocalrecoveryreserved_latency", "Wait local recovery reserved recovery state latency"); | |
| 304 | rs_perf.add_time_avg(rs_waitremoterecoveryreserved_latency, "waitremoterecoveryreserved_latency", "Wait remote recovery reserved recovery state latency"); | |
| 305 | rs_perf.add_time_avg(rs_recovering_latency, "recovering_latency", "Recovering recovery state latency"); | |
| 306 | rs_perf.add_time_avg(rs_recovered_latency, "recovered_latency", "Recovered recovery state latency"); | |
| 307 | rs_perf.add_time_avg(rs_clean_latency, "clean_latency", "Clean recovery state latency"); | |
| 308 | rs_perf.add_time_avg(rs_active_latency, "active_latency", "Active recovery state latency"); | |
| 309 | rs_perf.add_time_avg(rs_replicaactive_latency, "replicaactive_latency", "Replicaactive recovery state latency"); | |
| 310 | rs_perf.add_time_avg(rs_stray_latency, "stray_latency", "Stray recovery state latency"); | |
| 311 | rs_perf.add_time_avg(rs_getinfo_latency, "getinfo_latency", "Getinfo recovery state latency"); | |
| 312 | rs_perf.add_time_avg(rs_getlog_latency, "getlog_latency", "Getlog recovery state latency"); | |
| 313 | rs_perf.add_time_avg(rs_waitactingchange_latency, "waitactingchange_latency", "Waitactingchange recovery state latency"); | |
| 314 | rs_perf.add_time_avg(rs_incomplete_latency, "incomplete_latency", "Incomplete recovery state latency"); | |
| 315 | rs_perf.add_time_avg(rs_down_latency, "down_latency", "Down recovery state latency"); | |
| 316 | rs_perf.add_time_avg(rs_getmissing_latency, "getmissing_latency", "Getmissing recovery state latency"); | |
| 317 | rs_perf.add_time_avg(rs_waitupthru_latency, "waitupthru_latency", "Waitupthru recovery state latency"); | |
| 318 | rs_perf.add_time_avg(rs_notrecovering_latency, "notrecovering_latency", "Notrecovering recovery state latency"); | |
| 319 | ||
| 320 | return rs_perf.create_perf_counters(); | |
| 321 | } |